Model design method, data processing method, device, electronic equipment and medium

By introducing a hybrid encoder consisting of CNN and RNN concatenated into the Transformer model, the problem of high computational complexity of the Transformer model on embedded platforms is solved, achieving faster model inference speed and higher computational efficiency.

CN122197973APending Publication Date: 2026-06-12BEIJING AUTOMOBILE RES GENERAL INST
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
BEIJING AUTOMOBILE RES GENERAL INST
Filing Date
2026-01-19
Publication Date
2026-06-12

AI Technical Summary

⚠Technical Problem

Transformer models have high computational complexity and low computational efficiency on embedded platforms, making them difficult to compute effectively.

⚗Method used

The target model is constructed by replacing the encoder in the Transformer model with a hybrid encoder consisting of a concatenated convolutional neural network (CNN) and a recurrent neural network (RNN).

🎯Benefits of technology

This reduces the computational complexity of the model and improves computational efficiency, enabling the target model to be computed quickly on embedded platforms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
Patent Text Reader

Abstract

The present disclosure provides a model design method, a data processing method, an apparatus, an electronic device and a medium. The method comprises: obtaining an original Transformer model, the Transformer model comprising a decoder and an encoder; constructing a hybrid encoder, the hybrid encoder being composed of a convolutional neural network (CNN) and a recurrent neural network (RNN) in series; and replacing the encoder in the Transformer model with the hybrid encoder to obtain a target model. The technical solution provided by the embodiments of the present disclosure can reduce the computational complexity of the model and speed up the inference speed on the embedded end without affecting the recognition accuracy by structurally fusing the CNN, the RNN and the transformer model.
Need to check novelty before this filing date? Find Prior Art